1
$\begingroup$

I'm using LSTM to predict financial data. As input data I use log returns and I want to predict the next day market movement. Do I need to retrain the ANN every day in order to keep time consistency or I can simply train ANN once for example with the data from 2010 to 2018 and predict market movement in 2019?

I'm using daily data

$\endgroup$
2
  • $\begingroup$ Depends on your data. Are you using daily, monthly or yearly data? + your question need more details. $\endgroup$ – Dawny33 Mar 26 '19 at 10:54
  • $\begingroup$ I'm using daily data $\endgroup$ – Andrew Mar 26 '19 at 11:05
1
$\begingroup$

An interesting idea would be to train the model with data between 2010 and 2018 and then keep training it every day to keep it updated.

Interesting related works can be found here and here.

Anyway, you need to decide what you want to predict it: do you want a daily output, monthly or yearly?

$\endgroup$
1
  • $\begingroup$ thx, I need daily output $\endgroup$ – Andrew Mar 26 '19 at 11:25
0
$\begingroup$

If you want to predict daily prices, you should use daily prices for testing/training. The big unknown, is the frequency of analysis for this kind of thing. You can certainly do inter-day trades. Looking at daily returns will produce different results from weekly, monthly, etc. Are you actually doing daily trades? This will certainly generate more commissions and it will cause higher volatility in your returns as well. Monthly re-balancing seems much more attractive, unless the daily process (or inter-day process) is 100% automated and 100% fault-tolerant.

Take a look at the two links below for some ideas of how to do LSTM for stock analysis.

https://towardsdatascience.com/simple-stock-price-prediction-with-ml-in-python-learners-guide-to-ml-76896910e2ba

https://www.analyticsvidhya.com/blog/2018/10/predicting-stock-price-machine-learningnd-deep-learning-techniques-python/

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.